“Price rose in 2 of 4 similar episodes” describes a frequency in selected historical observations. It does not mean a 50% chance of a rise in the next episode. To interpret the count, you need the selection rule, the observation period and the size of the moves.
A complete example with four episodes
The following data are entirely hypothetical, for teaching purposes; they are not Scan Chart results or a backtest. Before inspecting the outcomes, fix the same observation period for every episode. Measure both the final price change and the lowest price reached during that period relative to its starting price.
| Episode | Final change | Lowest price relative to the start |
|---|---|---|
| A | +1% | −3% |
| B | +1% | −8% |
| C | −10% | −12% |
| D | −10% | −18% |
Two of four episodes finished higher: the observed frequency is 2 / 4 = 50%. The mean final change is (+1 + 1 − 10 − 10) / 4 = −4.5%. These are different summaries: counting positive endpoints gives each episode equal weight regardless of the size of its move.
Episode B finished at +1%, but its price first fell as low as −8% relative to the start. A positive endpoint therefore does not describe the risk along the way. The last column is a minimum relative to the starting price, not a peak-to-trough drawdown or the loss of a specified trading strategy. Entry, exit, costs and execution have not been modelled.
The illustration with two price paths is a separate example: both finish at +6%, but the paths carry different adverse moves. Those two paths are not rows from this table.
Sample description
- Rise: 50%; Mean: −4.5
- The same outcome share can hide different loss sizes.
Why the observation period matters
“Higher afterwards” is incomplete without a defined endpoint. A price can be higher at one endpoint and lower at another. Choose the period before examining outcomes, report it alongside the count and compare like with like. The same number of candles covers different amounts of time on different timeframes and trading calendars.
The sample length describes the historical shape being compared. The subsequent observation period answers a separate question. There is no universal inverse relationship between them.
What can make the sample misleading
Few observations. Four episodes form a small sample; changing a single episode changes the frequency substantially. The fraction alone gives no reliable estimate of a future event's probability.
Dependence. Nearby or overlapping episodes can reflect the same market event. Even distinct windows may share a market regime, so the number of rows need not equal the number of independent observations.
Selection and missing history. Results depend on which assets and periods have data. Omitting delisted assets, failed projects or difficult periods can bias a historical sample; the direction and size of the bias require checking.
Repeated choices. Trying many lengths, thresholds and endpoints, then keeping the most attractive result, fits the past. Record choices before seeing the subsequent moves and test the rule on a separate period.
How to read the result
Inspect individual paths, dispersion and adverse moves alongside the frequency and mean. A context label such as price relative to EMA200 can help describe differences between episodes, but it cannot make them interchangeable.
A useful historical comparison produces a question to test. It does not by itself establish a forecast or trading edge. The review of research on technical analysis explains what additional evidence a trading claim would need.
Path to outcome
- A
- B
- Outcome: +6%
- Drawdown: 0% / 10%
- The share of past outcomes is not the next outcome’s probability.